← All guides

How to generate usable portraits on Nanomelon

A working session plan for faces, crops, and light—not a vibe dump

Most failed portraits on this site are not “bad models.” They are unfinished briefs. A line like “a girl, with blonde hair, radiating sunshine” (see the public notes on gallery 153) forces the engine to invent age, wardrobe, camera distance, and whether sunshine is weather or a rim light. This guide is the portrait-specific version of our general prompt formula.

Name the person without naming a celebrity

Write an age band and one visual fact: “young woman, freckles, linen shirt” or “older man, grey stubble, wool coat.” Avoid real public figures and brand mascots; those prompts often die in the safety filter and teach you nothing. “Girl” without context can land as a child. If you mean an adult, say so. If you mean a child, keep the framing non-sexual and ordinary—Nanomelon is a general-audience tool and we will not help you push that line.

Expression belongs in the pose sentence: looking just off-camera, mid-laugh, serious. “Beautiful smile” is weaker than “closed-mouth smile, eyes toward camera left.”

Camera distance is the crop

Decide the crop before you romanticize the light.

On Nanomelon the aspect-ratio control is part of the prompt. Generating a 16:9 face and cropping later throws away the composition the model spent capacity on. The pixel sizes you see on gallery cards (for example 1024×1365) are tall on purpose—match them when you want a poster look.

Light: replace “sunshine” with a direction

Sunshine is not a setting. Useful substitutes:

Put warmth in objects (dry grass, brick, wood) rather than in a stack of “golden hour, magic hour, god rays.” Those stacks produce haze disks and fried highlights. If a result is too yellow, neutralize the grade in the next prompt instead of adding more sun adjectives.

Skin and finish

If you want a photograph, say “natural skin texture, visible pores, no beauty blur.” If you want illustration, name the medium once (gouache, ink, comic) so the model does not default to a beauty-filtered photo. Mixing “ultra detailed photo” with “anime eyes” usually gives neither. For the anime-versus-photo decision tree, use anime vs realistic.

A portrait session that spends credits on purpose

  1. Write subject + pose + place + light + medium. Do not generate yet.
  2. Pick one model. Set quantity to 1. Generate.
  3. Change only light or only camera distance. Generate once.
  4. When the face is close enough, switch to image-to-image and describe wardrobe or background only: “same person and crop, swap hill for balcony, keep backlight.”

Failed and policy-blocked jobs do not take credits; ugly successes do. Raising quantity before the brief is stable is how starter balances disappear. Cost context: models and credits.

Hands, hair, and glasses

Hands near the face fail often. Prefer “hands in pockets” or “hands out of frame” until you need them. Hair across the eyes is another lottery—say “hair tucked behind one ear” if you need a clean brow. Glasses: “thin metal frames, no glare on lenses” is clearer than “wearing glasses.”

Do not paste other people’s private photos into a public gallery work without permission. Public-visible toggles put the prompt in plain text for strangers.

Checklist before Generate

Open the generator with quantity 1. Compare a short mood prompt against the rewritten brief pattern on gallery 153.